Legal

Code of Ethics

The principles that govern how Paybacker uses generative AI, primary UK legal sources, and user data when drafting consumer complaint and dispute letters.

1. Overview

Paybacker drafts UK consumer complaint and dispute letters using generative AI grounded in primary legal sources. This document is the de facto code of ethics already operating inside the product — it describes commitments enforced in code, in process, or in published product behaviour, not aspirations. It corresponds to question 19 of the Find Case Law (TNA) Computational Analysis Agreement application.

2. Human-in-loop

AI proposes, the user approves. Paybacker never auto-sends a complaint letter, escalation request, cancellation email, or small-claims draft. The autonomous Dispute Agent surfaces recommended next actions but always requires a user click before any outbound communication is generated or sent. Human approval is the load-bearing safeguard against AI error.

3. Source attribution

Every citation in a Paybacker letter links back to the original primary source — caselaw.nationalarchives.gov.uk for judgments, legislation.gov.uk for statutes, the regulator's own domain for guidance. The user, and any recipient of the drafted letter, can always verify a citation against the canonical record.

4. No republishing of judgment text

Paybacker does not republish the full text of any Find Case Law judgment. Drafted letters quote only the neutral citation, the court, and a short plain-English ratio summary, with a back-link to the original record. Aggregate citation lists or extracted-data feeds are not published.

5. Plain-English explanation

Every formal legal reference in a Paybacker letter is paired with a plain-English explanation of what it means for the consumer's specific dispute. This is a deliberate accessibility choice: consumers should be able to read their own letter and understand it without legal training.

6. No automated terminal outcomes

Paybacker's AI never decides whether a dispute is “won” or “lost”. The dispute outcome dataset captures won / partial / lost / withdrawn / timeout / still-open states only when the user (or in some cases an AI proposal that the user confirms) explicitly records them. The engine surfaces a suggested outcome with an evidence excerpt; the user has to click Confirm. Terminal outcomes are never auto-written.

7. Founder-reviewed corrections queue

Citations drift — statutes are amended, regulator guidance is republished, judgments are superseded. Paybacker runs a daily compliance-sync pipeline that flags any citation whose source appears to have changed. Every proposed change is queued for founder review before the canonical record is updated. Semantic changes (section numbers, year changes, act renames, jurisdiction changes) always require founder approval; only same-host redirect fixes within the authority allowlist may auto-apply.

8. GDPR posture

User data — account information, dispute facts, uploaded correspondence — is held under the UK GDPR. Paybacker LTD is the data controller. We do not extract, store or republish personal data from Find Case Law judgments. Citations reference cases by neutral citation, court and ratio only — names of parties, judges and witnesses appearing in the public record are not used as analytical inputs and are not stored beyond the citation itself. Full details of how we handle user data, including UK GDPR rights, are set out in our Privacy Policy.

9. AI transparency

Paybacker's use of generative AI is disclosed explicitly in marketing pages and in the onboarding flow. Inside the product, every AI-drafted letter is presented to the user as an AI-assisted draft requiring their review before sending. The AI provider (Anthropic Claude) and the human-in-loop review step are described in our Methodology page.

10. Bias monitoring

We run a monthly sample audit of cited authorities to check that the engine is not over-relying on a narrow set of citations or skewing dispute outcomes by demographic factors. A quarterly summary records the audit findings, any corrections applied, and any changes to the relevance-scoring policy. Findings that would affect cited authorities feed directly into the corrections queue described in section 7.

11. Contact

Ethics questions, concerns, or proposed corrections: hello@paybacker.co.uk. Paybacker LTD is registered in England & Wales (company no. 15289174).